基于深度学习和空间关系推理的椎间盘定位标注方法  

Localization and labeling of intervertebral disc based on deep learning and spatial relation reasoning

作　　者：王生生[1] 邵婧雯 李妮娅[1] 刘纯岩[2] 李文辉[1] 陈鹏[2] 刘冰 WANG Sheng-sheng;SHAO Jing-wen;LI Ni-ya;LIU Chun-yan;LI Wen-hui;CHEN Peng;LIU Bing(College of Computer Science and Technology,Jilin University,Changchun 130012,China;The Second Hospital of Jilin University,Changchun 130041,China)

机构地区：[1]吉林大学计算机科学与技术学院,长春130012 [2]吉林大学第二医院,长春130041

出　　处：《吉林大学学报（工学版）》2020年第3期1085-1090,共6页Journal of Jilin University:Engineering and Technology Edition

基　　金：吉林省科技发展计划项目(20190302117GX,20180101334JC,20170204020GX);吉林省发展改革委创新能力建设(高技术产业部分)项目(2019C053-3).

摘　　要：为实现在核磁共振图像中对椎间盘自动定位标注,提出了基于深度学习算法和空间关系推理的椎间盘自动定位标注新方法。首先,利用深度学习算法对椎间盘进行检测,并采用评分约束提高算法精度。其次,给出邻近椎间盘空间关系模型,基于该模型匹配每个椎间盘区域的中心点。最后,采用空间关系迭代推理算法选择最佳匹配结果,得到椎间盘标注序列。本文假设每张核磁共振图像至少包含6个连续的椎间盘,实验结果表明:本文方法经过训练后测试效果良好,准确度能够达到临床要求,且执行效率高,有助于椎间盘疾病的计算机辅助诊断。In order to realize the automatic localization and labeling of intervertebral disc in Magnetic Resonance Imaging(MRI)images,a new method based on deep learning algorithm and spatial relation reasoning is proposed.First,the deep learning algorithm is applied to detect the intervertebral disc and scoring constraints are used to improve the accuracy of the algorithm.Second,the spatial relationship model of adjacent intervertebral discs is presented,based on which the center points of each intervertebral disc region are matched.Finally,the optimal matching results are selected by the spatial relation iterative reasoning algorithm,and the disc labeling sequence is obtained.It is assumed that each MRI image contains at least 6 consecutive intervertebral discs.The experimental results show that the method has a good test effect after training,the accuracy can meet the clinical requirements,and the execution efficiency is high,which is helpful for the computer-aided diagnosis of intervertebral disc disease.

关 键 词：计算机应用 椎间盘定位标注 深度学习 评分约束 空间关系推理 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]